I-PAttnGAN: An Image-Assisted Point Cloud Generation Method Based on Attention Generative Adversarial Network

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper introduces an image-assisted point cloud completion network, I-PAttnGAN, which effectively addresses the challenges of generating high-quality point clouds, particularly in sparse regions. While the paper demonstrates certain innovative contributions, further optimization, and revisions are required before it can be accepted.

1. The description of the innovations in the Introduction section could be reorganized for better clarity. It is suggested to first discuss the overall innovation at a high level and then elaborate on the contributions of individual modules, providing detailed explanations of the strategies employed in each component.

2. The diagram illustrating the overall structure in the paper (Figure 2.)appears relatively simplistic. For instance, the blue bold arrows represent the two distinct encoders and the "Attention Weighted" operation. It might be more appropriate to use different shapes or symbols with greater distinction to represent these elements more clearly.

3. In the "Main Idea" subsection (around line 156), the authors state: "In contrast to the majority of point cloud generation methods that focus solely on global representation, our approach caters to the requirements for both long-range object representation and local point cloud details, which are pivotal in tasks such as 3D map construction." To enhance the argument’s persuasiveness, it would be helpful to summarize how this dual representation is achieved briefly. Without this clarification, the statement risks appearing incomplete or lacking substantiation.

4. The paper would benefit from a more comprehensive explanation of the overall network structure in Chapter 3, as this would improve clarity and help readers better understand the model’s working mechanisms. For example, while the experimental section mentions that ResNet is used as the image encoder, this detail could be included in Chapter 3 to avoid leaving readers with unanswered questions.

5. In the experimental section, the comparison algorithms across the three sub-tasks only include PointFlow (2019), AtlasNet (2018), DEM (2021), and PointNet++ (2017). Including results from more recent models would make the experiments more convincing.

 

6. In the ablation study results, Figure 4 uses the term "Our_adaptive" directly. It might be clearer to provide an explicit explanation of this term either in the figure caption or in the accompanying text.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The authors in this paper innovatively propose an image-assisted point-cloud-generating method that can handle sparse regions well.  Below are some comments that need to be addressed to improve the quality and clarity of the manuscript:

1. The introduction lacks a clear logical structure and key elements, e.g., stat-of-the-art and challenges, are not expressed clearly.

2. There is a lack of detailed description of the related work on image-assisted point cloud generation.

3. Further, the classification and correlation of related work need to be reconstructed.

4. The description of the overall architecture is not comprehensive.

5. The symbols in the formulas lack detailed explanations.

6. The details of the method generation part are insufficiently described, and more explanation is needed.

7. The contribution of the proposed method is not expressed in a specific and clear manner.

8. The structure of the whole paper needs to be re-orchestrated to improve its coherence and readability.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

3D point clouds, a fundamental data model of environmental perception, have been widely adopted in various AI-powered scenarios, e.g., robots and autonomous vehicles. This paper investigated how to leverage image data to enhance 3D point clouds in sparse regions, thereby improving efficiency while minimizing resource waste.

Although the proposed method is innovative, several issues should be addressed before it can be considered for publication.

1.      The terms “CD” and “EMD” in the abstract should be replaced with their full names for clarity.

2.      The introduction section lacks a summary of related work on image-assisted approaches, and the challenging problem to be solved in this paper is not clearly highlighted.

3.      Section 3 lacks a comprehensive description of the overall architecture associated with the proposed method.

4.      It is suggested to add a preliminary section to detail the procedures for building 3D point clouds and the backgrounds of technical cornerstones related to this work.

5.      The experimental setup subsection does not provide a clear high-level overview of the evaluation questions; it is not clear what the purposes of 4.2, 4.3, and 4.4 are.

6.      In page. 9. it is stated that “these experiments significantly improved our proposed method on the synthetic dataset ShapeNet.” What does this mean? Or a typo?

7.      PointFlow [29] is used as a baseline in the experimental section, but it is not discussed in the related work.

Author Response

Please see the attachment.

Author Response File: Author Response.docx